Source code for torcheval.metrics.functional.classification.confusion_matrix
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
from typing import Optional
import torch
norm = torch.nn.functional.normalize
[docs]@torch.inference_mode()
def binary_confusion_matrix(
input: torch.Tensor,
target: torch.Tensor,
*,
threshold: float = 0.5,
normalize: Optional[str] = None,
) -> torch.Tensor:
"""
Compute binary confusion matrix, a 2 by 2 tensor with counts ( (true positive, false negative) , (false positive, true negative) )
See also :func:`multiclass_confusion_matrix <torcheval.metrics.functional.multiclass_confusion_matrix>`
Args:
input (Tensor): Tensor of label predictions with shape of (n_sample,).
``torch.where(input < threshold, 0, 1)`` will be applied to the input.
target (Tensor): Tensor of ground truth labels with shape of (n_sample,).
threshold (float, default 0.5): Threshold for converting input into predicted labels for each sample.
``torch.where(input < threshold, 0, 1)`` will be applied to the ``input``.
normalize:
- ``None`` [default]:
Give raw counts ('none' also defaults to this)
- ``'pred'``:
Normalize across the prediction, i.e. such that the rows add to one.
- ``'true'``:
Normalize across the condition positive, i.e. such that the columns add to one.
- ``'all'``"
Normalize across all examples, i.e. such that all matrix entries add to one.
Examples::
>>> import torch
>>> from torcheval.metrics.functional import binary_confusion_matrix
>>> input = torch.tensor([0, 1, 0.7, 0.6])
>>> target = torch.tensor([0, 1, 1, 0])
>>> binary_confusion_matrix(input, target)
tensor([[1, 1],
[0, 2]])
>>> input = torch.tensor([1, 1, 0, 0])
>>> target = torch.tensor([0, 1, 1, 1])
>>> binary_confusion_matrix(input, target, threshold=1)
tensor([[0, 1],
[2, 1]])
>>> input = torch.tensor([1, 1, 0, 0])
>>> target = torch.tensor([0, 1, 1, 1])
>>> binary_confusion_matrix(input, target, normalize="true")
tensor([[0.0000, 1.0000],
[0.6667, 0.3333]])
"""
_confusion_matrix_param_check(2, normalize)
matrix = _binary_confusion_matrix_update(input, target, threshold)
return _confusion_matrix_compute(matrix, normalize)
[docs]@torch.inference_mode()
def multiclass_confusion_matrix(
input: torch.Tensor,
target: torch.Tensor,
num_classes: int,
*,
normalize: Optional[str] = None,
) -> torch.Tensor:
"""
Compute multi-class confusion matrix, a matrix of dimension num_classes x num_classes where each element at position `(i,j)` is the number of examples with true class `i` that were predicted to be class `j`.
See also :func:`binary_confusion_matrix <torcheval.metrics.functional.binary_confusion_matrix>`
Args:
input (Tensor): Tensor of label predictions.
It could be the predicted labels, with shape of (n_sample, ).
It could also be probabilities or logits with shape of (n_sample, n_class).
``torch.argmax`` will be used to convert input into predicted labels.
target (Tensor): Tensor of ground truth labels with shape of (n_sample, ).
num_classes (int):
Number of classes.
normalize:
- ``None`` [default]:
Give raw counts ('none' also defaults to this)
- ``'pred'``:
Normalize across the prediction class, i.e. such that the rows add to one.
- ``'true'``:
Normalize across the condition positive, i.e. such that the columns add to one.
- ``'all'``"
Normalize across all examples, i.e. such that all matrix entries add to one.
Examples::
>>> import torch
>>> from torcheval.metrics.functional import multiclass_confusion_matrix
>>> input = torch.tensor([0, 2, 1, 3])
>>> target = torch.tensor([0, 1, 2, 3])
>>> multiclass_confusion_matrix(input, target, 4)
tensor([[1, 0, 0, 0],
[0, 0, 1, 0],
[0, 1, 0, 0],
[0, 0, 0, 1]])
>>> input = torch.tensor([0, 0, 1, 1, 1])
>>> target = torch.tensor([0, 0, 0, 0, 1])
>>> multiclass_confusion_matrix(input, target, 2)
tensor([[2, 2],
[0, 1]])
>>> input = torch.tensor([0, 0, 1, 1, 1, 2, 1, 2])
>>> target = torch.tensor([2, 0, 2, 0, 1, 2, 1, 0])
>>> multiclass_confusion_matrix(input, target, 3)
tensor([[1, 1, 1],
[0, 2, 0],
[1, 1, 1]])
>>> input = torch.tensor([0, 0, 1, 1, 1, 2, 1, 2])
>>> target = torch.tensor([2, 0, 2, 0, 1, 2, 1, 0])
>>> multiclass_confusion_matrix(input, target, 3, normalize="pred")
tensor([[0.5000, 0.2500, 0.5000],
[0.0000, 0.5000, 0.0000],
[0.5000, 0.2500, 0.5000]])
>>> input = torch.tensor([0, 0, 1, 1, 1])
>>> target = torch.tensor([0, 0, 0, 0, 1])
>>> multiclass_confusion_matrix(input, target, 4)
tensor([[2, 2, 0, 0],
[0, 1, 0, 0],
[0, 0, 0, 0],
[0, 0, 0, 0]])
>>> input = torch.tensor([[0.9, 0.1, 0, 0], [0.1, 0.2, 0.4, 0.3], [0, 1.0, 0, 0], [0, 0, 0.2, 0.8]])
>>> target = torch.tensor([0, 1, 2, 3])
>>> multiclass_confusion_matrix(input, target, 4)
tensor([[1, 0, 0, 0],
[0, 0, 1, 0],
[0, 1, 0, 0],
[0, 0, 0, 1]])
"""
_confusion_matrix_param_check(num_classes, normalize)
sparse_cm = _confusion_matrix_update(input, target, num_classes)
return _confusion_matrix_compute(sparse_cm, normalize=normalize)
def _binary_confusion_matrix_compute(
cm: torch.Tensor, normalize: Optional[str]
) -> torch.Tensor:
if normalize == "pred":
return norm(cm.to(torch.float), p=1, dim=1)
elif normalize == "true":
return norm(cm.to(torch.float), p=1, dim=0)
elif normalize == "all":
return cm.to(torch.float) / torch.sum(cm)
else:
return cm
def _binary_confusion_matrix_update(
input: torch.Tensor,
target: torch.Tensor,
threshold: float = 0.5,
) -> torch.Tensor:
_binary_confusion_matrix_update_input_check(input, target)
input = torch.where(input < threshold, 0, 1)
return _update(input, target, 2)
def _binary_confusion_matrix_update_input_check(
input: torch.Tensor,
target: torch.Tensor,
) -> None:
if input.ndim != 1:
raise ValueError(
f"input should be a one-dimensional tensor for binary confusion matrix, got shape {input.shape}."
)
if target.ndim != 1:
raise ValueError(
f"target should be a one-dimensional tensor for binary confusion matrix, got shape {target.shape}."
)
if input.shape != target.shape:
raise ValueError(
"The `input` and `target` should have the same dimensions, "
f"got shapes {input.shape} and {target.shape}."
)
def _confusion_matrix_compute(
confusion_matrix: torch.Tensor,
normalize: Optional[str],
) -> torch.Tensor:
if normalize == "pred":
return norm(confusion_matrix.to(torch.float), p=1, dim=0)
elif normalize == "true":
return norm(confusion_matrix.to(torch.float), p=1, dim=1)
elif normalize == "all":
return confusion_matrix.to(torch.float) / torch.sum(confusion_matrix)
else:
return confusion_matrix
def _confusion_matrix_update(
input: torch.Tensor, target: torch.Tensor, num_classes: int
) -> torch.Tensor:
_confusion_matrix_update_input_check(input, target, num_classes)
return _update(input, target, num_classes)
@torch.jit.script
def _update(
input: torch.Tensor,
target: torch.Tensor,
num_classes: int,
) -> torch.Tensor:
if input.ndim == 2:
input = torch.argmax(input, dim=1)
coordinates = torch.vstack((target, input))
cm_shape = torch.Size([num_classes, num_classes])
# Each prediction creates an entry at the position (true, pred)
sparse_cm = torch.sparse_coo_tensor(coordinates, torch.ones_like(target), cm_shape)
return sparse_cm.to_dense()
def _confusion_matrix_param_check(
num_classes: int,
normalize: Optional[str],
) -> None:
if num_classes < 2:
raise ValueError("Must be at least two classes for confusion matrix")
if (normalize is not None) and (normalize not in ["all", "pred", "true", "none"]):
raise ValueError("normalize must be one of 'all', 'pred', 'true', or 'none'.")
def _confusion_matrix_update_input_check(
input: torch.Tensor,
target: torch.Tensor,
num_classes: Optional[int],
) -> None:
if input.size(0) != target.size(0):
raise ValueError(
"The `input` and `target` should have the same first dimension, "
f"got shapes {input.shape} and {target.shape}."
)
if target.ndim != 1:
raise ValueError(
"target should be a one-dimensional tensor, " f"got shape {target.shape}."
)
# check if num classes is high enough to cover inputs.
if not input.ndim == 1:
if not (input.ndim == 2 and (input.shape[1] == num_classes)):
raise ValueError(
f"input should have shape of (num_sample,) or (num_sample, num_classes), "
f"got {input.shape}."
)
else:
if torch.max(input) >= num_classes:
raise ValueError(
"Got `input` prediction class which is too large for the number of classes, "
f"num_classes: {num_classes} must be strictly greater than max class predicted: {torch.max(input)}."
)
# check if num classes is high enough to cover targets.
if torch.max(target) >= num_classes:
raise ValueError(
"Got `target` class which is larger than the number of classes, "
f"num_classes: {num_classes} must be strictly greater than max target: {torch.max(target)}."
)
